Robotics
RoboBugBench: Benchmarking Automated Repair of Real Robot Software Under an Audited Behavioral Oracle
Alexey Tuzikov
Language agents increasingly modify
robot software, but whether they can repair it is unmeasured, because the domain supplies no oracle to inherit: a
robot bug manifests as behavior, and
robot repositories ship almost no tests that capture it. We present RoboBugBench, a
benchmark of 33 real, execution-validated defects in a pinned ROS 2
navigation stack, in which a candidate patch is graded by compiling it and driving the
robot through a physics simulation rather than by running a unit test. Three frontier models, each given 1 attempt per instance, repair 24.2%-36.4%, with no pairwise difference significant; difficulty tracks the oracle rather than the domain, at 66.7% on instances graded by process survival against 22.2% on those graded by what the
robot did (instance-level p = 0.0027). Because the oracle is constructed rather than inherited we audit it adversarially, and report the decomposition that audit yields: counterfeit patches which suppress a failure without repairing the defect are accepted by the shipped rule 11/11, while a distributional rule over the same data rejects 4 at no cost in false rejects. The residual is an observability bound (what the observation channel discards, no predicate over it can recover), so discrimination must be bought with instrumentation rather than stricter criteria, and our repair rates are bounds: inflated by what the oracle cannot distinguish, deflated by unanimity's rejection of correct fixes under run-to-run noise. We release the corpus, criterion batteries, counterfeit suite and cached telemetry.
Read the full abstract ▾
Machine Learning
LoopQuant: Accurate and Efficient Post-Training Quantization for Looped Transformers
Zhixiong Zhao, Zukang Xu, Xing Hu et al. (+3 more)
Looped
Language Models (LoopLMs) improve parameter efficiency by recursively reusing shared
Transformer modules, yet their repeated computation over evolving recurrent states poses new challenges for low-bit deployment. Existing post-training quantization (PTQ) methods, largely designed for feed-forward
Transformers, fail to account for cross-loop activation dynamics, heterogeneous loop sensitivity, and recursive error propagation, leading to substantial performance degradation at low precision. To address these challenges, we propose LoopQuant, a loop-aware PTQ framework for accurate and efficient low-bit quantization of LoopLMs. LoopQuant is built on three core components: 1) Cross-Loop Distribution Canonicalization (CLDC) learns a shared quantization-friendly representation across recurrent states while preserving parameter sharing; 2) Interaction-Aware Loop-wise Precision Allocation (ILPA) allocates activation precision under a fixed bit budget by modeling task-level loop sensitivity and cross-loop interactions; and 3) Loop-Conditioned Task Calibration (LCTC) further aligns quantized inference with the low-bit recursive trajectory through lightweight task-space calibration. Extensive experiments across LoopLM architectures demonstrate that LoopQuant consistently outperforms existing low-bit PTQ methods. On Ouro-2.6B, LoopQuant under W4A4 surpasses all W4A8 baselines in performance, while reducing model
memory by 65.3% with only 3.4% additional storage overhead over naive INT4 and achieving a 1.7 times decoding speedup over BF16. These results demonstrate a strong balance among accuracy, compression, and inference efficiency, providing a practical and deployment-friendly solution for low-bit LoopLM inference. The code is available at [https://github.com/Kishon-zzx/LoopQuant].
Read the full abstract ▾
All authors Zhixiong Zhao, Zukang Xu, Xing Hu, Lifeng Liu, Guangyu Sun, Dawei Yang
Soft Condensed Matter
Derjaguin-corrected Kelvin equation for finite open slits
Alexandr Malijevský
We formulate a Derjaguin-corrected Kelvin equation for phase transitions in finite open slits. In such pores, capillary condensation is modified by the finite wall length because the condensed phase is bounded by two terminal menisci pinned at the slit ends. Starting from the macroscopic generalized Kelvin equation, we introduce a scaling approximation which replaces the implicit edge-contact-angle condition by a closed analytic expression while preserving the exact limiting behaviours. We then incorporate mesoscopic wetting-layer contributions into the finite-slit free-energy balance. The single-wall contribution naturally separates into a planar Derjaguin term and a residual finite-length correction controlled by the competition between the wall length $H$ and the parallel correlation length $\xi_\parallel$. The resulting equation reduces to the standard Derjaguin-Kelvin equation for infinitely long slits and to the macroscopic finite-slit Kelvin equation when adsorbed layers are absent. Explicit forms are derived for long-ranged dispersion forces and for short-ranged forces. The theory is tested using classical density functional theory for finite slits with attractive Lennard-Jones walls and hard walls, corresponding respectively to capillary condensation and its drying analogue, capillary evaporation. The Derjaguin-corrected theory substantially improves on the purely macroscopic prediction, capturing both the correct infinite-slit limit and the observed dependence on wall length. For the short-ranged hard-wall system, the agreement with DFT remains nearly quantitative even for very narrow slits.
Read the full abstract ▾
Classical Physics
Experiments on an airfoil with adjustable pitch and heave under transonic buffet leading to lock-in coupling
Cédric Themiot, Vincent Brion and Arnaud Lepage
This paper presents aeroelastic experiments conducted in the ONERA S3Ch transonic wind tunnel using a supercritical OAT15A airfoil mounted on a rig allowing independent control of pitch and heave degrees of freedom (DOF). The study investigates fluid-structure interaction (FSI) mechanisms and lock-in phenomena between transonic buffet and structural modes. Aerodynamic and structural responses were monitored using steady and unsteady pressure sensors, high-speed stereoscopic Particle Image Velocimetry (PIV), accelerometers, and optical sensors. Four configurations were tested: rigid (0 DOF), pitch-only (1 DOF pitch), heave-only (1 DOF heave), and combined pitch-heave (2 DOF). The test campaign spanned selected Mach numbers, angles of attack, and structural natural frequencies for each configuration. Results show that buffet-induced lock-in occurs abruptly when the Mach number increases, but only when the pitch DOF is enabled, in both the 1-DOF pitch or 2-DOF configuration. These findings establish that pitch-buffet frequency proximity is a necessary condition for lock-in. In the 2-DOF configuration, continuous Mach number excursions revealed a three-way coupling involving the buffet, pitch, and bending modes, with the bending mode modulating the coupled dynamics. This coupling generates large-amplitude limit-cycle oscillations (LCO) within seconds, persisting over a narrow Mach range before buffet cessation.
Read the full abstract ▾
Artificial Intelligence
World Models Need a Physical Registry: Learning Internal Predictive States for Physically Plausible World Models
Shinseong Kim, Jaeseok Jeong, Jianlong Fu et al. (+1 more)
Video world models must generate physically plausible futures, not just realistic frames. Existing approaches typically use semantic, physical, or predictive information to guide the visual generation stream, leaving a fundamental question: where inside the generator should predictions of scene
evolution live? Standard
video tokens jointly encode appearance and dynamics, but provide no distinct state for learning and reading predictive dynamics. We answer this question with the Physical Registry: temporally grounded registry slots inside a
video DiT that are initialized from an observation-conditioned forecast, co-evolve with
video tokens, and receive future-representation supervision, making predictive dynamics an explicit, readable state of the generator. In a controlled comparison of predictive-state interfaces on Physics-IQ Verified, Physical Registry improves vanilla fine-tuning by 9.9/8.7 points on I2V/V2V, the largest gain among the interfaces we compare. Across all Cosmos and WAN backbones, Physical Registry improves Physics-IQ Verified scores. With only 0.45% additional parameters, it enables Cosmos 3 Nano 16B to surpass the published I2V score of Cosmos 3 Super 64B (43.56 vs. 42.7). On WorldModelBench, it also improves physical adherence and the total score for all four evaluated backbones. Because the registry becomes candidate-specific while remaining readable, the Physical Registry Selector (PRS) ranks emerging futures by how far their intermediate registry readouts depart from the shared forecast before full denoising is complete. Across four backbones, this selection improves I2V scores by 5.6-15.9% without completed-
video external scoring. Together, these results support a broader principle: dynamics prediction should live as an explicit model state that co-evolves with the
video representation. Project page: https://grainsack.github.io/physical-registry/
Read the full abstract ▾
All authors Shinseong Kim, Jaeseok Jeong, Jianlong Fu, Youngjung Uh
Computer Vision
From Sound to Sense: An Anatomy of Semantic Supervision for Speech-Driven Visual Grounding in Remote Sensing Imagery
Zhongyu Pang
Speech-driven visual grounding in remote sensing (RSVG) removes the typing bottleneck of text-based interaction and is intended to preserve prosodic intent, enabling hands-free target acquisition in time-critical missions. The direct-audio paradigm, mapping raw
speech straight to a bounding box and bypassing automatic
speech recognition (ASR), was recently adopted by RSTalker, but what such grounders actually learn about
language remains unmeasured. SemEcho grafts a frozen Whisper
speech pathway onto a VLTVG grounding host so that
speech and text share one decoder, making the paired text, free by construction since
speech corpora in this task are synthesized from DIOR-RSVG expressions, a training-time teacher and a fair, same-host text upper bound. On this instrument we dissect three semantic injection channels (feature-mimicry
distillation, word-level CTC supervision, and two-stage ASR pretraining) with linear semantic probes, and release Audio-DIOR-R (80,973 training clips; 10 timbres x K=3; Qwen3-TTS synthesis with online sox noise lambda in [0, 0.4]) and Audio-DIOR-Comp, a re-split holding out 17 attribute-category combinations under a 2x2 acoustic-semantic protocol. Our anatomy yields four findings. (i) Whether direct
alignment beats a domain-trained ASR cascade is decided by recognizer quality: against a weak in-domain cascade (CTC 0.53) direct wins by 9.0 Acc@0.5 points with transcription errors concentrated on spatial-relation words, but a strong in-domain cascade (CTC 0.029) reaches the text upper bound (80.8 vs. 81.0) and beats even the strongest direct model by 8.8 points; the anti-ASR premise holds only against weak recognizers.
Read the full abstract ▾
Information Theory
Log Density Spectrum Based Decoding of Polar Codes for Channels with Strong Phase Noise
Yinuo Mei and Daiming Qu
A decoding method is proposed for polar codes under channels with strong phase noise. In contrast to traditional SC decoding, where recursions are performed on scalar soft information such as probabilities, the proposed decoding recurses over probability density functions defined on the phase-noise variable. To make the density-level recursion computationally tractable, we represent each log-density by its spectrum, and derive approximate formulas for the spectrum recursion, therefore, the proposed method is termed log-density spectrum recursion (LDSR). Numerical results are given under strong phase noise and AWGN, showing that a prominent superiority of error performance is achieved by LDSR in comparison with traditional SC and Kalman-based methods.
Read the full abstract ▾
Representation Theory
Double Bruhat cells in symmetric spaces
Huanchen Bao and Jinfeng Song
This paper studies the closures of the double Bruhat
cells of symmetric spacess in its equivariant
embeddings. We call them large double Bruhat varieties. Our aim is to establish the double Bruhat stratification as a natural setting for extending the geometry of Richardson varieties to symmetric spaces. For the wonderful compactification of the associated adjoint symmetric space, we show their scheme-theoretic intersections with the unique closed orbit are reduced unions of projected Richardson varieties. This connection with Richardson geometry on partial flag varieties supports large double Bruhat varieties as a natural class of subvarieties in symmetric-space
embeddings. In positive characteristic, we prove that every equivariant
embedding admits a Frobenius splitting compatible with all large double Bruhat varieties and all boundary divisors. These varieties are strongly F-regular, and globally F-regular when the ambient
embedding is projective. In characteristic zero, they are of strongly F-regular type, and of globally F-regular type in the projective case; in particular, they are normal, Cohen-Macaulay, and have rational singularities.
Read the full abstract ▾
Computation and Language
HexaBee-4B: Six-Language Translation with Multilingual Answering and Bidirectional Supervision
Iman Barati, Mohammadmostafa Rostamkhani and Mohammad Badzohreh
Multilingual translation requires a model to preserve meaning while producing text in a requested
language. We present HexaBee-4B, a translator adapted from Gemma 3 4B IT through three stages: Wikipedia continued pretraining, joint
multilingual answering and answer-text translation, and bidirectional translation specialization. The intermediate stage connects two tasks through shared
multilingual answers: answering a question in the requested
language and translating an answer into that
language. It comprises 19.46 million answering records, including 5.76 million cross-
language question-answer pairs, and 10.70 million answer-text translations. The final stage uses 658,274 forward and reverse synthetic translation examples. On 36,572 synthetic test examples spanning all 30 directions among Arabic, Chinese, English, Hebrew, Persian, and Spanish, HexaBee-4B achieves direction-macro chrF++ of 73.75, compared with 57.11 for Gemma 3 4B IT and 66.83 for Gemma 3 12B IT. BLEU, TER, COMET, and a matched model-judged subset support the same overall ranking. These results demonstrate improved translation across the six-
language set, with chrF++ gains in every evaluated direction over both baselines.
Read the full abstract ▾
Computational Geometry
Shaving Two Logs for Discrete and Continuous Fréchet Distance
Timothy M Chan and Isaac M Hair
We revisit the classical computational geometry problem of computing the exact (discrete or continuous) Fréchet distance between two given polygonal curves with $n$ vertices in a constant dimension $d$. 1. We present an $O((n^2/\log^2n)\log^2\log n)$-time deterministic
algorithm for the discrete Fréchet distance problem, improving the previous $O((n^2/\log n)\log\log n)$ result by Agarwal, Ben Avraham, Kaplan, and Sharir [SODA'13]. 2. We present an $O((n^2/\log^2n)\log^2\log n)$-time deterministic
algorithm for the decision version of the continuous Fréchet distance problem, improving the previous $O((n^2/\log n)\log^2\log n)$ result by Buchin, Buchin, Meulemans, and Mulzer [SODA'14] and the recent $O(n^2/\log^{1+\mu}n)$ randomized result by Cheng and Huang [SODA'25] for a small unspecified constant $\mu>0$ (which converges to 0 as $d$ increases). 3. Finally, we present an $O((n^2/\log^2 n)\log^{O(1)}\log n)$-time Las Vegas randomized
algorithm for the continuous Fréchet distance problem, improving the previous $O(n^2\log^2\log n)$ randomized result by Buchin, Buchin, Meulemans, and Mulzer [SODA'14] and the recent $O(n^2/\log^{\mu}n)$-time randomized result claimed by Cheng and Huang [SODA'25]. Since these problems have near-quadratic OV-based conditional lower bounds and shaving of two log factors is the best that is known for similar, simpler problems such as longest common subsequence and edit distance, our results reach the natural limit of the current state of the art (up to $\log\log$ factors).
Read the full abstract ▾
Numerical Analysis
An algebraic multiscale framework for two-level overlapping Schwarz preconditioners
Filipe Cumaru, Hadi Hajibeygi, Cornelis Vuik et al. (+1 more)
The robustness of two-level overlapping additive Schwarz (OAS) preconditioners for highly heterogeneous
elliptic problems hinges on the choice of the coarse space. We propose an algebraic multiscale (AM) coarse space built from the system
matrix alone through an algebraic formulation of the multiscale finite element method (MsFEM). The coarse basis functions are energy-minimizing extensions of the solutions of reduced boundary condition (RBC) problems on the subdomain interface, so that coefficient variations along the interface enter the coarse space. By posing the RBC problems on slabs that extend the interface, the construction also applies to general unstructured domain decompositions.
Embedding the AM formulation in an energy-minimizing framework extends it to linear elasticity, and we show that existing condition number bounds for OAS with MsFEM coarse spaces carry over to the algebraic setting. For problems with high coefficient inclusions on the coarse nodes, the AM iteration count and condition number are unaffected by increasing the contrast. In 3D scalar
diffusion with log-normal coefficients, AM needs about as many iterations as the generalized Dryja–Smith–Widlund (GDSW) coarse space, while solving a coarse problem ten times smaller. For linear elasticity, omitting the coarse basis functions associated with the rotational rigid body modes yields a fully algebraic coarse space that still scales well for homogeneous problems.
Read the full abstract ▾
All authors Filipe Cumaru, Hadi Hajibeygi, Cornelis Vuik, Alexander Heinlein
Computational Complexity
NP-hardness of additive approximation for generalized Turán problems
Yihan Chen, Jialin He, Xihe Li et al. (+2 more)
For a fixed graph $T$ and a fixed nonempty family $\mathcal{F}$ of graphs, let $\operatorname{ex}(G, T, \mathcal{F})$ denote the maximum number of copies of $T$ in an $\mathcal{F}$-free subgraph of a graph $G$. Alon and Shikhelman~[Algorithmica, 2022] conjectured that if $e(T)> 0$ and no member of $\mathcal{F}$ is a subgraph of a blow-up of $T$, then for every $\varepsilon>0$, approximating $\operatorname{ex}(G, T, \mathcal{F})$ within an additive error $n^{v(T)-\varepsilon}$ is NP-hard for an input graph $G$ on $n$ vertices. The previously established general cases for this
conjecture include $\operatorname{ex}(G, K_2, \mathcal{F})$ by Alon, Shapira and Sudakov~[Ann. of Math., 2009], and $\operatorname{ex}(G, K_m, K_k)$ by Alon and Shikhelman. In this paper, we confirm this
conjecture for every bipartite graph $T$ and every clique $T$. Our reductions use iterative leaf additions, symmetric rooted extensions and random transversals to encode the maximum number of edges in a $(k-1)$-colorable subgraph of $G$, where $k=\min_{F\in \mathcal{F}}\chi(F)$.
Read the full abstract ▾
All authors Yihan Chen, Jialin He, Xihe Li, Jie Ma, Tianying Xie
Computer Vision
VideoSlate: Stateful Commit Control for Temporally Grounded Agentic Video Reasoning
Shuo Zheng, Yuanjie Dang, Peng Chen et al. (+4 more)
Agentic
video reasoning systems gather and assess
video evidence across stages, selecting relevant spans as visual evidence for answering questions. New candidates and observations
prompt reassessment of the current span, but revisions can correct localization errors or discard useful evidence already present in the adopted span. Systems therefore need to connect evidence organization across stages with temporal state control, allowing accumulated information to jointly inform revision assessment while explicit adoption decisions govern how assessment outcomes affect downstream visual evidence. We introduce VideoSlate, a stateful commit-control framework built around shared evidence. A typed evidence blackboard links candidate spans, provenance, construction, and verifier observations, enabling an incumbent-relative certifier to assess revision support from the candidate, current span, and related proposals. Based on certification, the controller authorizes a revision or retains the current span, separating evidence accumulation from state updates and recording each decision and its evidence on the blackboard. Extensive evaluations across
video grounding and grounded question-answering
benchmarks show that VideoSlate can improve temporal localization through selective revision.
Read the full abstract ▾
All authors Shuo Zheng, Yuanjie Dang, Peng Chen, Jianjun Xiang, Ronghua Liang, Mohammed Bennamoun, Qiuhong Ke
Neural and Evolutionary Computing
naspaclust: A hybridisation of spatial fuzzy clustering with nature-inspired optimisation
Bahrul Ilmi Nasution, Rezzy Eko Caraka and Robert Kurniawan
We present an overview of R package called naspaclust, a tool for spatial fuzzy clustering improvement by hybridising the Fuzzy Geographically Weighted Clustering (FGWC) with nature-inspired optimisation
algorithms. There are seven optimisation
algorithms in general, namely Artificial Bee Colony, Flower Pollination
Algorithm, Gravitational Search
Algorithm, Harris-Hawk Optimisation, Intelligent Firefly
Algorithm, Particle Swarm Optimisation, and Teaching-Learning Based optimisation. The optimisation
algorithms available do not only use the basic
algorithms, but also the modified version of several parts such as randomisation terms and updating rules. Furthermore, we also provide additional options by providing more distance metrics options based on the rdist package. This study showed the implementation of the naspaclust package in various cases, as well as comparing the performance of the
algorithms in the scope of basic
algorithms. The naspaclust package allows to improve the performance of fuzzy clustering, mainly in spatial context.
Read the full abstract ▾
Robotics
PINT: A Unified Model with Shared Predictive Intent for Robot Manipulation
Xinyu Guo, Chao Jing, Li Yi et al. (+2 more)
Generalist
robot manipulation can benefit from pretrained
vision-
language models (VLMs) for semantic
reasoning and
video generation models (VGMs) for world modeling. However, existing unified models that directly fuse them through shared
attention at multiple layers can make it harder to draw on their specialized capabilities. We introduce PINT, a unified model that coordinates semantic
reasoning, world modeling, and action generation through a shared chunk-level Predictive Intent while preserving expert-specific pathways. PINT inserts learnable Intent
Tokens into the VLM sequence to form this representation, conditioning both the VGM and an Action Expert (AE). It replaces only the VGM's
prompt conditioning and retains the VGM at inference, where the AE reads its representations through one-way
attention. Rather than prescribing what Predictive Intent should encode, PINT learns it jointly from the native prediction objectives of the VLM, VGM, and AE, which provide complementary views of the same upcoming task execution. Training uses two stages: Stage 1 freezes the VGM so that future
video prediction shapes Predictive Intent with task-relevant future dynamics, while Stage 2 fixes the learned Intent and adapts the VGM and AE. Without
embodied pretraining, PINT achieves strong in-distribution (ID) performance and substantially improves out-of-distribution (OOD) generalization, reaching 15.5% on RoboTwin 2.0-Hard compared with 1.8% for Fast-WAM and 3.2% for StarVLA, and 79.5% on LIBERO-Plus compared with 51.5% and 75.0%, respectively. Across five real-world bimanual tasks, PINT reaches 68% ID and 24% OOD success, compared with 35%/9% for pretrained Motus and 67%/35% for pretrained $\pi_{0.5}$.
Read the full abstract ▾
All authors Xinyu Guo, Chao Jing, Li Yi, Xingyu Chen, Zhengxing Wu
Computation and Language
Self-Distillation with Verbal Regularization
Jiazheng Li, Yawei Li, Junru Lu et al. (+7 more)
On-policy self-
distillation (OPSD) trains a large
language model using supervision from a teacher instantiated from the same model as the student. The teacher is conditioned on privileged information (PI), such as verified solutions. The student, which does not observe PI, learns from the teacher's next-
token distributions along its own sampled responses. However, PI can bias supervision toward shortcut
reasoning with limited verification or reconsideration, accompanied by a sharp drop in the
entropy of the output distribution. To address this challenge, we introduce self-
distillation with verbal regularization (VRSD), which uses system
prompts to regulate how the teacher uses PI, thereby encouraging rethinking and exploration to counteract premature commitment. Theoretically, we model prompting as attenuating the influence of PI on the teacher distribution and derive conditions for
entropy preservation. Across three models and four tasks, VRSD improves Pass@1 over OPSD in 11 of 12 model-task settings, by 3.9 percentage points on average, while generally maintaining higher
token entropy.
Read the full abstract ▾
All authors Jiazheng Li, Yawei Li, Junru Lu, Zhenyi Shen, Di Yin, Lin Gui, Xing Sun, David Rügamer, Xinpeng Wang, Yulan He
Combinatorics
The $(km,kn)$-square paths theorem
Dun Qiu and Minhao Zhang
For positive integers $k,m,n$ with $\gcd(m,n)=1$, we prove a $(km,kn)$-square paths
theorem: the symmetric function $(-1)^{k-1}\frac{[km]_q}{[k]_q}p_k[-MX^{m,n}]\cdot1$ is a weighted sum over $(km,kn)$-preference functions. This proves a corrected form of the
conjecture of Iraci, Pagaria, Paolini, and Vanden Wyngaerd and generalizes the square paths
theorem of Sergel. We encode preference functions by row strict tableaux and closed walks on a set of states, compare them with parking functions for each value of the touch statistic, and apply Mellit's compositional $(km,kn)$-shuffle
theorem. Finally, we establish a fall Delta square
theorem conjectured by Iraci, Pagaria, and Paolini.
Read the full abstract ▾
Machine Learning
LoongSage: A Production-Grade, High-Efficiency Agentic RL Framework for Frontier LLMs with Validated Recipes
LoongSage Team: Bowei Song, Caiwei Zhao, Chuwen Liu et al. (+44 more)
Reinforcement learning (RL) has emerged as a critical phase in the post-training pipeline of large
language models (LLMs). As the frontier of LLM capabilities shifts from static
reasoning to agentic tasks–characterized by external tool invocation, multi-turn interactions, and long-horizon decision-making–existing mainstream RL frameworks have grown increasingly unwieldy, raising the overhead of adoption, usage, and debugging. To address these challenges, we present LoongSage, an
agent-native, fully asynchronous RL framework with a lightweight core of roughly 14K lines of code. Built upon Megatron for distributed training and SGLang for high-throughput inference, LoongSage centralizes orchestration within a single controller process and encapsulates all GPU-resident components as Ray actors. It introduces AgentFlow, a unified execution paradigm that orchestrates both white-box and black-box
agents. For asynchronous execution, it integrates sliding-window strategies that bound trajectory staleness, and employs TransferMesh, a
topology-aware synchronization mechanism that routes weights after each step over the fastest transport available to each receiver. Beyond agentic RL, LoongSage incorporates a
memory-efficient Teacher Manager that decouples the scaling of teacher models from GPU
memory constraints, natively supporting Multi-Teacher On-Policy
Distillation. Extensive evaluations across four complete training runs, spanning complex reinforcement learning and multi-teacher
distillation, demonstrate that task rewards improve steadily while the train-inference gap is kept under continuous control, and the fully asynchronous pipeline removes synchronous idle bubbles, sustaining hardware utilization.
Read the full abstract ▾
All authors LoongSage Team: Bowei Song, Caiwei Zhao, Chuwen Liu, Dechuang Gu, Dong Fang, Feiyu Xie, Huaiyu Li, Hui Zhang, Jian Tian, Jian Wang, Jie Zhao, Liguang Li, Maoyun Wang, Meiyan Yang, Minghan Zhu, Pai Liu, Pandeng Yao, Renbao Tian, Ru Yang, Shuai Jia, Shuai Tan, Shuailong Li, Shuliang Li, Teng Jin, Tianze Yang, Wenbo Cui, Wenjing Bai, Wenkang Wu, Xiaowei Jiang, Xiaoyi Dong, Yanpeng Wang, Yanwen Wang, Yaohui Ma, Yawen Wang, Yihang Zhang, Yingjie Yan, Zhanbiao Lian, Zhenyu Lu, Zhida Hu, Zhihui Hao, Zhihui Hu, Zhiwen Xiao, Zhiyi Gao, Zhongwei Qi, Zhuangzhuang Wu, Ziqian Liu, Ziyue Zheng
Quantum Physics
Finite-time quantum equilibration from local correlations
Piotr Ćwikliński
An isolated
quantum system evolves reversibly, yet a small subsystem can lose nearly all information about its initial state. Conserved quantities may retain some of this information, depending on how strongly they overlap with local observables. Here we consider a subsystem in an arbitrary initial state coupled to a maximally mixed bath. We show that the worst-input mean-square equilibration error, maximised over window origins, is bounded from above and below by the mean square of a complete local autocorrelation, with the stationary part subtracted. A correlation record over a finite interval therefore controls the equilibration error in every interval of the same length. The result allows arbitrary degeneracies of energies and gaps. We apply it to two static
Hamiltonian families. For fixed
spin subsystems in the one-particle sector of a bounded-range periodic clock, fixed-accuracy thermalisation is guaranteed on windows that grow only slightly faster than linearly with system size, despite exponentially small positive energies. A random commuting model with logarithmic interaction range retains an extensive set of conserved observables. Its expected mean-square error is bounded by a finite-size term and an inverse-time term. The inverse-time dependence is unavoidable because of the initial transient. These examples show how local relaxation can coexist with exact conservation laws and a finely spaced spectrum.
Read the full abstract ▾
Algebraic Geometry
Generalizing Green's Syzygy Theorems to Symmetric Products of Curves
Wenbo Niu and Peter Yi Wei
Let $C$ be a smooth projective curve of genus $g$, and let $T_{n,L}$ denote the tautological line bundle on its $n$-th symmetric product $C_n$. We prove effective vanishing criteria for Koszul
cohomology $K_{p,q}(C_n,T_{n,B};T_{n,L})$, extending two classical
theorems of Mark Green. The main ingredient is a weight-one vanishing
theorem proved by induction on $n$, using a filtration of the kernel bundle of $T_{n,L}$ to reduce the required cohomological vanishings to lower symmetric products. As a consequence, when $g\geq1$ and $n\geq2$, $T_{n,L}$ satisfies property $(N_p)$ for $\mathrm{deg}(L)\geq(n+1)g+p$. This bound is sharp for
elliptic symmetric squares and also gives evidence for a folklore
conjecture on adjoint syzygies.
Read the full abstract ▾
Computation and Language
RIFT: Token Influence Follows the Path of Least Resistance
Tingan Jin, Shuhang Dong and Haosong Li
Which context
tokens drive a
language model's answer? Graph-based tools such as
attention rollout and
attention flow answer this from a single forward pass, yet their faithfulness in decoder-only models is rarely tested against interventions. We view the layers of a decoder as a causal network whose conductances are
attention weights, and read a
token's influence on the answer from the resistance between them. We propose RIFT (Resistance-based Information Flow in
Transformers), which scores a
token by the series resistance of its path of least resistance to the answer, found exactly by a classic shortest-path dynamic program. On synthetic in-context
retrieval tasks and extractive question answering over Wikipedia paragraphs, RIFT predicts the effect of replacing a
token with a Spearman correlation of 0.52-0.68; predicting a single replacement, it reaches 77-96% of the agreement between two different random replacements. It needs no gradients and no extra forward passes, scoring every
token of a
prompt in 0.9-3.4 ms on one CPU thread, whereas the replacement experiment needs a forward pass per
token. On 576 held-out
prompts from four models of two families (Qwen3 and Llama 3.2), it outperforms
token distance,
attention rollout and
attention flow in every pre-specified comparison, by 0.23 to 0.88 in mean Spearman correlation, and it is the best or tied-best of all
attention-based scores we test. Aggregates that add up all routes (
attention flow, effective resistance and rollout) instead mostly encode
token position: as a simple proposition shows, early
tokens have many more parallel routes to the answer.
Read the full abstract ▾
Machine Learning
DWI Connectome-based Sex Classification in Young using Supervised Variational Autoencoder
Aditya Kothapalli, Aaryan Gera, Sathvik Reddy V et al. (+3 more)
During adolescence, significant changes in
brain development are influenced by hormonal changes that impact both the body and the
brain. This study presents sex classification in 787 young adults having age range of 15-25 years using
Diffusion-weighted (DWI) structural connectome, focusing on 82 Desikan-Killiany atlas ROIs. A supervised variational autoencoder (SVAE) is used to extract latent space features from DWI connectome data. Although several studies explored differences in
brain structure and function between sexes, the main contribution of this study lies in incorporating SVAE to learn a compact, regularized latent representation for the task of sex classification. Initial exploratory data analysis shows that these features can capture the differences in sex-specific latent space patterns. Subsequently, the learned latent space SVAE features are fed to different ML models for sex classification. Among the considered ML classifiers, the multi-layer perceptron yields the highest accuracy of 80% in classifying sex. This demonstrates the effectiveness of SVAE in learning compact, informative representations from DWI connectome for sex classification. The present work highlights the potential of using DWI
brain connectomics and deep learning to differentiate sex, paving the way for future research on diverse populations that may include transgender individuals, to better understand sex-related variations in
brain structure.
Read the full abstract ▾
All authors Aditya Kothapalli, Aaryan Gera, Sathvik Reddy V, Ninad Aithal, Debanjali Bhattacharya, Neelam Sinha
Machine Learning
Rescuing Points Trapped Inside Incorrect Decision Regions: A target for Evaluating and Improving Forward-Forward Learning
Alireza Rahmansetayesh, Amirhossein Yousefiramandi, Maxwell Spiridonov et al. (+3 more)
Forward-Forward learning replaces end-to-end backpropagation with local, layer-wise training objectives. Its variants are evaluated primarily by classification accuracy, often measured with linear probes trained on their representations. We show that this evaluation can over-credit the learning method: a linear probe on the representations of an untrained convolutional neural network already reaches accuracy well above chance. Moreover, a single accuracy value does not reveal which datapoints become correctly classified after training and which remain misclassified. To examine what learning contributes beyond the random, untrained baseline, we present an analysis methodology built on the Boundary Proximity Index (BPI), a multiclass margin normalized by a class-centroid reference. We apply our analysis to seven Forward-Forward variants and related layer-local methods and find that initially misclassified points near the probe's decision boundary generally have higher rescue rates than those farther inside incorrect decision regions. Rescue rates vary substantially across methods, with some correcting a substantial fraction of these deeply embedded points. Together, these analyses offer a way to evaluate representation learning beyond accuracy and identify the rescue of points far inside incorrect decision regions as a target for improving local learning methods.
Read the full abstract ▾
All authors Alireza Rahmansetayesh, Amirhossein Yousefiramandi, Maxwell Spiridonov, Sumit Raut, Kent Ziti, Pierre Baldi
Databases
CellChain: Benchmarking Cross-Scale Evidence Use for Perturbation Reasoning
Xue Xia, Yue Wang, Shuizhou Chen et al. (+6 more)
Recent advances in foundation models and large-scale cellular perturbation
datasets have driven growing interest in using LLMs to predict and reason about cellular responses. Biological
reasoning in this setting involves relating molecular mechanisms, cellular responses, and functional outcomes across scales. Evaluating foundation models therefore requires assessing not only whether their predictions are correct, but also how those predictions depend on the biological evidence provided. We introduce CellChain, an experimentally grounded
benchmark comprising 590 biological cases and 2,360 matched evidence-condition instances across gene-dependency prediction, cytokine-axis identification, and drug-sensitivity ranking. CellChain links perturbation experiments and external biological resources through a provenance-aware evidence graph spanning molecular, pathway, transcriptional, and phenotypic information. Task-specific conditions vary evidence coverage, composition, or assignment while preserving questions, candidate answers, reference targets, and scoring rules. Across 18 foundation models, the highest accuracies under prespecified target-evidence conditions are 57.5%, 75.5%, and 70.8%, respectively, and broader evidence coverage does not consistently improve accuracy. On the drug-sensitivity
benchmark cohort, matched reference histories outperform swapped histories by an average of 13.0 percentage points in cluster-weighted accuracy. Decision-transition analysis separates corrections, newly introduced errors, and wrong-answer switches, revealing substantial answer changes hidden by small net accuracy differences. CellChain provides controlled comparisons for attributing changes in model predictions to prespecified expansions of biological context, including changes in evidence coverage, composition, and biological assignment.
Read the full abstract ▾
All authors Xue Xia, Yue Wang, Shuizhou Chen, Lang Yu, Hao Wu, Wenxuan Huang, Chengkai Yao, Zhangyang Gao, Siqi Sun
Artificial Intelligence
ChronoScribe: Semantic-Numerical Specialization for Instruction-Driven Time Series Forecasting
Yuchen Jiao, Miao Zhang, Yuxuan Liu et al. (+1 more)
Recent advances in instruction-driven time series
forecasting highlight a fundamental tension: systems must simultaneously master natural-
language instruction following and precise numerical prediction. Current approaches often force a compromise. Large
language models (LLMs) can interpret unstructured requests but may be inaccurate when asked to produce forecast values directly, whereas time series foundation models (TSFMs) provide numerical forecasts but typically require structured inputs specifying the target series, context window, and horizon. We present ChronoScribe, which applies semantic-numerical specialization to this workflow. A LoRA-adapted LLM proposes an execution path and context length; deterministic rules resolve the executable configuration. On the specialist path, a frozen TSFM produces
quantile trajectories for the task-specified horizon. Deterministic trajectory summaries condition the response, while the LLM neither generates nor revises predicted values. We evaluate a Qwen3.5-9B–Chronos-2 implementation across 23 evaluation dimensions and 30 public
datasets against eleven proprietary direct LLMs and three open TSFMs. ChronoScribe has the lowest error on sMAPE, MASE, WAPE, RMSE, and MAE among all evaluated systems, including fixed-context controls. Across these metrics, even the lowest direct-LLM error remains 2.0–2.7 times the corresponding ChronoScribe error. At the system level, ChronoScribe achieves a Task-Level Score (TLS) of 65.5, 10.6 points above the strongest direct-LLM baseline. The results support ChronoScribe's division of labor, in which the LLM handles interpretation, configuration, and communication while the TSFM handles numerical
forecasting.
Read the full abstract ▾
All authors Yuchen Jiao, Miao Zhang, Yuxuan Liu, Chengxi Li
Human-Computer Interaction
STEPwise: Supporting Teachers in Diagnosing Students' Mathematical Problem-Solving Skills through LLM-Generated Questions and Rubric
Yoonsu Kim, Hayeon Doh, Eunhye Kim et al. (+2 more)
Problem-solving is central to mathematics education. However, diagnosing where students'
reasoning breaks down requires individualized probing and consistent criteria that teachers cannot sustain at classroom scale. We introduce STEP (STage-based Evaluation of Problem-solving skills), co-designed with six mathematics education experts, which operationalizes problem-solving as four stages and eight diagnostic sub-skills. By specifying what evidence to elicit and assess at each stage, STEP enables an LLM pipeline to generate stage-specific questions and rubrics for any math word problem, reducing the manual design of diagnostic materials. An expert review (N=8) assessed the pedagogical validity of the generated materials and informed STEPwise, a teacher-in-the-loop system for diagnostic material generation and process-level assessment. In a classroom study with 5 teachers and 64 students, STEPwise helped teachers and students locate
reasoning breakdowns and supported students in reflecting on their
reasoning. Together, STEP and STEPwise offer a structured approach to fine-grained diagnosis at classroom scale.
Read the full abstract ▾
All authors Yoonsu Kim, Hayeon Doh, Eunhye Kim, Jinho Son, Juho Kim
Artificial Intelligence
Efficient Multi-Agent Reasoning via Confidence-Guided Adaptive Debate
Seungdong Yoa, Ye Seul Sim, Suhee Yoon et al. (+5 more)
Multi-
agent debate has shown promise for improving the
reasoning of large
language models, yet recent theory suggests its benefits are highly regime-dependent. While interaction can amplify informative signals under corrective conditions, symmetric debate dynamics are neutral in expectation, often making majority voting preferable. We reconcile these views by arguing that debate is effective only when invoked at the right time and with appropriate structure. Based on this insight, we propose LASE: Leader-Adaptive Structured Engagement, a leader-centric multi-
agent debate framework that selectively engages interaction only in non-neutral regimes. LASE introduces an asymmetric leader-supporter architecture that enables directed information flow and selective signal amplification, while defaulting to simple aggregation otherwise. Experiments across diverse
reasoning benchmarks show that LASE achieves multi-
agent-level performance with near single-
agent token cost, substantially improving efficiency over static debate and voting baselines.
Read the full abstract ▾
All authors Seungdong Yoa, Ye Seul Sim, Suhee Yoon, Sanghyu Yoon, Dongmin Kim, Soonyoung Lee, Bumsoo Kim, Junhyun Lee
Signal Processing
Analysis of Integrated Communication and Sensing in Cylindrical Aerial Corridors
Harris K. Armeniakos, Petros S. Bithas, Athanasios G. Kanatas et al. (+1 more)
In unmanned aerial vehicle (
UAV) networks, integrated communication and sensing (
ISAC) is emerging as a key enabler to support extended and continuous communication and sensing capabilities for sixth-generation (6G) services among
UAVs operating in swarms. In this paper, a comprehensive framework is presented to investigate the
ISAC coverage probability of a reference
UAV-base station (BS) in a
UAV network. A fixed number of
UAV user equipment (UAVUEs) and
UAV-BSs are deployed in an aerial corridor modeled as a finite cylinder. The three-dimensional (3D) binomial point process (BPP) is used to model both the 3D spatial locations of the
UAV-UEs and the
UAV-BSs. By adopting realistic 3D third generation partnership project (3GPP)
antenna patterns for both the
UAV-UEs and
UAV-BSs, the
UAVs' boresight direction of the maximum
antenna gain is modeled as a random unit vector. All
UAV-BSs have sensing capabilities, so an arbitrary yet fixed reference
UAV-BS acts as a monostatic radar to acquire sensing measurements associated with a target
UAV-UE at a given time snapshot while the target UAVUE communicates with the reference
UAV-BS. Consequently, considering both clutter and communication interference, exact analytical expressions for
ISAC coverage probability are derived. Low complexity baseline analytical expressions are also derived by ignoring clutter interference and directional characteristics of the
UAV-UEs. Finally, motivated by the
vision for intelligent
UAV networks, a DeepSets-based model is developed and trained to estimate the optimal location of the reference
UAV-BS that maximizes
ISAC performance given a spatial deployment of the
UAVs. To the best of our knowledge, this is the first work to rigorously investigate
ISAC performance in
UAV networks while incorporating the directionality characteristics of the
UAV antennas.
Read the full abstract ▾
All authors Harris K. Armeniakos, Petros S. Bithas, Athanasios G. Kanatas, Harpreet S. Dhillon
Analysis of PDEs
One-dimensional pressureless flow with a nonlinear velocity law: Bregman variational construction and well-posedness
Gaowei Cao, Feimin Huang and Guirong Tang
We establish global existence, uniqueness and quantitative stability for one-dimensional pressureless flow governed by \[ \partial_t\rho+\partial_x\bigl(\rho f(u)\bigr)=0, \qquad \partial_t(\rho u)+\partial_x\bigl(\rho u f(u)\bigr)=0, \] where $\rho$ is the mass density, possibly measure-valued, $u$ is the physical velocity, and $f\in C^1(\mathbb R)$, with $f'>0$, determines the transport velocity $f(u)$. The initial mass has finite positive mass and finite second moment, allowing atoms and unbounded support, and the initial physical velocity is bounded. The difficulty at a cluster collision is that mass and momentum conservation determine a mass-weighted average of physical velocity, whereas transport is governed by the nonlinear law $f$ applied to that average. We construct global
entropy solutions by a Bregman variational time discretization whose cost is chosen to reproduce the transport velocity determined by this mass-weighted average, coupled with conditional-expectation updates of physical velocity. For every
entropy solution satisfying the Oleinik condition on the transport velocity and initial weak continuity of the kinetic energy measure, we establish a Lagrangian representation and prove uniqueness within this class by comparing flows with possibly different cluster intervals. For two such solutions with the same velocity law and common mass, we obtain a $W_\infty$ estimate for the normalized mass measures in terms of the initial position and physical-velocity discrepancies in mass coordinates, whenever the initial $W_\infty$ distance is finite. The Lagrangian
evolution has the
semigroup property.
Read the full abstract ▾
Classical Analysis and ODEs
Translational self-affine tiles in $\mathbb{R}^n$
Li-Xiang An and Ting-Ting He
Two basic problems in the theory of integral self-affine tiles are to characterize the digit sets that generate translational tiles and to understand whether such tiles admit periodic translational tilings. We give a complete solution to the first problem: for an expanding integer
matrix $\mathbf{A}$ and a digit set $\mathcal{D}\subset\mathbb{Z}^n$ with $\#\mathcal{D}=|\det\mathbf{A}|$, the self-affine set $T(\mathbf{A},\mathcal{D})$ is a tile if and only if, for some $N\geq1$, the iterated digit set $\mathcal{D}_{{\mathbf{A}},N}$ is of skew-product form with respect to $\mathbf{A}^N$. As a consequence, every integral self-affine tile admits a fully periodic self-replicating $q$-fold tiling. Moreover, $q$ divides $|T(\mathbf{A},\mathcal{D})|$. In dimension two we prove more: every integral self-affine tile admits a fully periodic tiling. In particular, the periodic tiling
conjecture holds for integral self-affine tiles in $\mathbb{R}^2$.
Read the full abstract ▾